megachad cua

It doesn't think.
It knows.

A voice-to-action API for computer-use agents. Your user talks, the agent moves before they finish the sentence. 15 ms per decision on a Qwen 3.8 27B trained to decide, not to chat.

Get your key Talk to it live

Checking the model…

Example

Model update ~15 ms Word to decision ~150 ms Chain of thought 0 tokens

Today's PRs.

0Calls served today

–Fastest decision today

3GPU slots open right now

0Reasoning tokens. Ever.

Live from the model. Resets at midnight UTC.

Average agent.
Chad agent.

The average agent

Screenshots the page, then thinks about it for a while
Waits for you to stop talking
Writes paragraphs of reasoning to click "Next"
"Just to confirm, did you mean…"
Bills you for every token it writes back

The chad agent

Reads the buttons. Decides in 15 ms.
Moves mid-sentence.
One decision. Zero reasoning tokens.
Changes its mind when you do.
6¢ a minute. Done.

Copy. Paste. Done.

Send the choices on screen as a schema. Stream the mic. Get back only what changed. Your key goes where it says mcc_live_….

# Text in, decision out.
curl https://api.megachadcua.com/v1/decide \
  -H "Authorization: Bearer mcc_live_…" \
  -H "content-type: application/json" \
  -d '{"schema": {"action": {"question": "What should the agent do?",
       "options": ["Click Sign in", "Click Pricing", "Scroll down", "Nothing yet"]}},
       "text": "take me to sign in"}'

# {"fields": {"action": {"value": "Click Sign in", "confidence": 0.98}}, "model_ms": 15, "latency_ms": 96}

# pip install "websockets>=14"   ·   voice in, decisions out while they talk
import asyncio, json, websockets

KEY = "mcc_live_…"
SCHEMA = {"action": {"question": "What should the agent do?",
                     "options": ["Click Sign in", "Click Pricing", "Scroll down", "Nothing yet"]}}

async def run(mic):  # mic: async iterator of 16 kHz mono PCM16 chunks
    async with websockets.connect("wss://api.megachadcua.com/v1/listen",
                                  additional_headers={"Authorization": f"Bearer {KEY}"}) as ws:
        await ws.send(json.dumps({"type": "start", "schema": SCHEMA,
                                  "audio": {"encoding": "pcm_s16le", "sample_rate": 16000}}))
        async def read():
            async for m in ws:
                ev = json.loads(m)
                if ev["type"] == "fields": print(ev["fields"])
        reader = asyncio.create_task(read())
        async for chunk in mic: await ws.send(chunk)
        await ws.send(json.dumps({"type": "end"})); await reader

// Browser: stream the mic, act when it's sure.
const ws = new WebSocket("wss://api.megachadcua.com/v1/listen", ["megachad", "mcc_live_…"]);
ws.onopen = () => ws.send(JSON.stringify({
  type: "start",
  schema: { action: { question: "What should the agent do?",
                      options: ["Click Sign in", "Click Pricing", "Scroll down", "Nothing yet"] } },
  audio: { encoding: "pcm_s16le", sample_rate: 16000 },
}));
ws.onmessage = (e) => {
  const ev = JSON.parse(e.data);
  if (ev.type === "fields" && ev.fields.action?.confidence > 0.9) agent.do(ev.fields.action.value);
};
mic.onframe = (pcm16) => ws.send(pcm16.buffer);  // Int16Array at 16 kHz from an AudioWorklet

# pip install "megachad[mic]"   ·   export MEGACHAD_API_KEY=mcc_live_…
import megachad
for f in megachad.listen_mic({"action": {"question": "What should the agent do?", "options": ["Scroll down", "Go back"]}}):
    print(f)  # action='Go back' (0.97)

# JS: npm install megachad   ·   await listenMic({ key, schema, onFields })

Pricing.

6¢per minuteVoice. /v1/listen

10¢per 1,000 decisionsText. /v1/decide

First $2 free, about 33 minutes of voice or 20,000 decisions. No card to start.
Voice bills per second, from the moment the model is ready.
Silence still counts while a session is open. Idle sockets close after 60 s.
Bigger text calls (over ~4,000 tokens) count as more than one decision.
No seats. No tiers. No sales call.
Stripe invoices you monthly. Cancel whenever.

Questions.

Is it fast?: Yes.
Can it handle "no wait, the other one"?: Yes.The answer flips the moment the words change.
Does it work with my agent?: Yes.If it can read the screen, send the buttons as options.
Text instead of voice?: Yes.One POST to /v1/decide.
Do I need a card to start?: No.$2 free.